Scribd
Upload
27 views3 pages

Block Size For Local Estimation

Local Estimation.

Uploaded by

Fred
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
27 views3 pages

Block Size For Local Estimation

Local Estimation.

Uploaded by

Fred
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 3

Block Size for Local Estimation

"While it is well known that a smoothing estimation method such as kriging is not
suitable to produce realistic estimates in blocks which are far away from the sampling
information, it is unfortunately too often used for this purpose." Ravenscroft and
Armstrong (1990).

Linear estimation, particularly that based on very widely spaced data relative to the size
of block to be estimated, is subject to two competing effects, conditional bias and
smoothing (Pan (1998)). Of these, smoothing imposes limits on the size of block that can
be estimated effectively because sparse, widely spaced data produce comparable
estimates for small adjacent blocks that do not reflect the true local variability of small
block grades.

Local reserve estimation at the feasibility stage commonly utilizes widely spaced
exploration data as a basis for estimating average grades of relatively small blocks. This
undertaking is almost too easy with the ready access to kriging software for personal
computers because it can lead to substantial errors if estimates are made for blocks whose
dimensions are small relative to sample spacing.

Armstrong and Champigny (1989) address the question, "How small (a block size) is too
small?" They examine variations in the kriging variance, dispersion of estimated grades,
slope of the regression of the true but unknown grades on estimated values and the
coefficient of correlation between actual grades and estimates. These four quantities were
calculated for several simple, 2-D geometries of data versus block using a spherical
semivariogram model with a sill of 1.0, no nugget effect and a range varying from 0.1m
to 40m (twice the sample spacing). Important generalities from their study are:

The kriging variance for block estimates changes significantly as a function of


block size and range of the semivariogram.
The kriging variance differs fundamentally from the dispersion variance; the two
should not be confused.
The dispersion variance of block grades varies widely depending on block size
and the range of the semivariogram.
Slope of the regression of actual grades Z on the estimated grades Zk* and the
correlation coefficient should both approach 1.0. Results show this to be the case
only if the semivariogram range is at least as great as the average sample spacing.
Where the range is small relative to sample spacing small block estimates are
"virtually useless for distinguishing ore from waste" and "should not be used for
predicting recoverable reserves".
Improvements in quality of block estimates can be achieved by using increased
amounts of surrounding data for block estimations if the semivariogram range is
close to the sample spacing (the difference in kriging error from adding a few
fringing data is minor).
Where the range is greater than 1.5 times the sample spacing, block estimates are
adequate to distinguish ore from waste.
These results by Armstrong and Champigny (1989) lead to the following guidelines for
successful small block kriging:

Data spacing should be substantially less than the range of the semivariogram, perhaps no
more than two thirds the range, and block dimensions should be as large as possible and
not much smaller than the average data spacing. A rule of thumb is that each block to be
estimated should contain at least one sample (Ravenscroft and Armstrong (1990)).

Ravenscroft and Armstrong (1990) document an example in which 10m-spaced data,


extracted from a set of real data spaced at 2m, are used to estimate 2m blocks. Kriged
estimates of 2m blocks using all available data are assumed to represent the "real" block
values to which estimated values are compared. In this case they show that for cutoff
values below the mean, kriging overestimates tonnage; for cutoff values above the mean,
kriging underestimates tonnage. At relatively high cutoffs, kriging estimates 5% of
blocks as ore instead of the 12% known to be ore! Moreover, kriging consistently
underestimates average grade above cutoff, regardless of cutoff value. Ravenscroft and
Armstrong (1990) conclude that

"...kriging of small blocks from sparse data will always over-estimate the recoverable
tonnage for a cut-off value below the mean, and under-estimate this tonnage for a cut-off
higher than the mean. The method will also always under-estimate recovered mean
grade, whatever the cut-off value."

An important consequence of the foregoing results is that kriging generally is not a sound
basis for the direct estimation of individual selection mining units from widely spaced
exploration data. Instead, a variety of techniques have been developed to determine
"recoverable" reserves from relatively limited exploration data. One of the general
approaches to this problem involves obtaining a kriging estimate of a large panel and a
separate estimate of the dispersion of small block (smu) grades within this large panel.
This dispersion of block grades is then centered on the kriged mean to provide an
estimate of the distribution of small block (smu) grades. Of course, the locations of these
individual, ore-grade, smu's within the large panel are not known, nor is the form of the
probability density function of the smu grades; assumptions are necessary.
A serious difficulty with many applications of computerised block modelling is the tendency for
uninformed practitioners to estimate very small blocks (in comparison to the spacing of available drilling).
This is the ‘small block linear estimate problem’ referred to previously. While it would be ideal to estimate
blocks at the scale of the envisaged selective mining unit (SMU) dimensions at the stage of feasibility, this
is rarely possible in practice. As block size decreases relative to drill spacing the precision of individual
block estimates decreases, often sharply. The grade-tonnage curves implicit for such estimates will be
distorted and conditionally biased (in other words, application of a cut-off grade greater than zero will
inevitably result in an incorrect estimation of tonnage and grades), and as a consequence, mine planning
based on these estimates (for example, pit optimisations) may be seriously economically misleading. This
applies to both ordinary kriging and inverse distance weighting methods (Ravenscroft and Armstrong,
1990; Krige, 1997). Stephenson, P R and Vann, J, 2001.

It is important to understand that the block size is critical in all cases where a cut-off will be applied to an
estimate (ie unless the estimate is to be used only in a global sense and without a cut-off). There is a long
bibliography of warnings against estimation of small blocks (for example Armstrong and Champigny,
1989; Ravenscroft and Armstrong, 1990; Royle, 1979; Vann and Guibal, 2000). The question is, how small
is too small?

You might also like